# DeepSeek-V2 (MoE-236B, May 2024) vs Llama 3-70B

> Llama 3-70B has enough public results to be ranked (#323); DeepSeek-V2 (MoE-236B, May 2024) does not yet, so treat this comparison as directional.

- Canonical page: https://noometry.com/compare/deepseek-v2-vs-llama-3-70b
- Last updated: 2026-10-10
- Shared benchmarks: 5

## Summary

- They share 5 benchmarks with published results for both. DeepSeek-V2 (MoE-236B, May 2024) scores higher in 1 category and Llama 3-70B in 0 categories; one gap is clear of the uncertainty.
- The widest gap is in coding, where DeepSeek-V2 (MoE-236B, May 2024) leads 40.4 to 35.8.
- The biggest single-benchmark swing is BigCodeBench Instruct: 48.9% for DeepSeek-V2 (MoE-236B, May 2024) and 43.6% for Llama 3-70B.

## Snapshot

| | DeepSeek-V2 (MoE-236B, May 2024) | Llama 3-70B |
|---|---|---|
| Provider | DeepSeek | Meta |
| Noometry Index | 40.3 | 28.8 |
| Rank | — | 323 |
| Context | — | — |
| Input $/M | — | — |
| Output $/M | — | — |
| Weights | Open | Open |

## Coding

- DeepSeek-V2 (MoE-236B, May 2024): 40.4 (#139)
- Llama 3-70B: 35.8 (#218)

| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | Llama 3-70B |
|---|---|---|
| BigCodeBench Instruct | 48.9% | 43.6% |
| BigCodeBench Complete | 59.4% | 54.5% |
| LMArena Coding | — | 1206 |
| HumanEval+ | — | 72% |
| MBPP+ | — | 69% |

## Agentic & Tool Use

- DeepSeek-V2 (MoE-236B, May 2024): —
- Llama 3-70B: 21.1 (#139)

| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | Llama 3-70B |
|---|---|---|
| Cybench | — | 5% |

## Reasoning

- DeepSeek-V2 (MoE-236B, May 2024): —
- Llama 3-70B: 18.0 (#288)

| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | Llama 3-70B |
|---|---|---|
| Epoch Capabilities Index | 124.77 | 122.93 |
| WinoGrande | 86.3% | 83.5% |
| Kagi LLM Benchmark | — | 35.1% |
| LMArena Hard Prompts | — | 1195 |
| DTBench | — | 54.2% |
| BIG-Bench Hard | 78.8% | — |
| ForecastBench | — | 57.1 |
| HellaSwag | 87.1% | — |
| PIQA | 83.9% | — |

## Math

- DeepSeek-V2 (MoE-236B, May 2024): —
- Llama 3-70B: 12.8 (#305)

| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | Llama 3-70B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | — | 4.3% |
| LMArena Math | — | 1218 |
| MATH Level 5 | — | 22.6% |

## Knowledge

- DeepSeek-V2 (MoE-236B, May 2024): —
- Llama 3-70B: 20.8 (#277)

| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | Llama 3-70B |
|---|---|---|
| MMLU | 78.4% | 79.3% |
| GPQA Diamond | — | 40.6% |
| LMArena Expert | — | 1149 |
| ARC (AI2) Challenge | 92.2% | — |
| TriviaQA | 80% | — |

## Multilingual

- DeepSeek-V2 (MoE-236B, May 2024): —
- Llama 3-70B: 33.6 (#251)

| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | Llama 3-70B |
|---|---|---|
| LMArena Non-English | — | 1142 |
| LMArena Chinese | — | 1114 |
| LMArena French | — | 1232 |
| LMArena German | — | 1169 |
| LMArena Japanese | — | 1017 |
| LMArena Korean | — | 1017 |
| LMArena Russian | — | 1159 |
| LMArena Spanish | — | 1241 |

## Instruction Following

- DeepSeek-V2 (MoE-236B, May 2024): —
- Llama 3-70B: 62.5 (#238)

| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | Llama 3-70B |
|---|---|---|
| LMArena Instruction Following | — | 1194 |

## Long Context

- DeepSeek-V2 (MoE-236B, May 2024): —
- Llama 3-70B: 35.6 (#240)

| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | Llama 3-70B |
|---|---|---|
| LMArena Longer Query | — | 1174 |

## Writing & Preference

- DeepSeek-V2 (MoE-236B, May 2024): —
- Llama 3-70B: 42.8 (#231)

| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | Llama 3-70B |
|---|---|---|
| LMArena Text | — | 1221 |
| LMArena Creative Writing | — | 1210 |
| LMArena Multi-Turn | — | 1223 |

## FAQ

### Is DeepSeek-V2 (MoE-236B, May 2024) better than Llama 3-70B?

Llama 3-70B has enough public results to be ranked (#323); DeepSeek-V2 (MoE-236B, May 2024) does not yet, so treat this comparison as directional.

### Is DeepSeek-V2 (MoE-236B, May 2024) or Llama 3-70B better for coding?

DeepSeek-V2 (MoE-236B, May 2024) scores higher on coding benchmarks: 40.4 versus 35.8 in the Noometry coding category.

### How many benchmarks do DeepSeek-V2 (MoE-236B, May 2024) and Llama 3-70B share?

5 benchmarks have published results for both models. DeepSeek-V2 (MoE-236B, May 2024) has 10 scored results on Noometry and Llama 3-70B has 31.